I’m really excited to try the FaceTrack CHOP/SOP in 12380. I tried a simple setup just trying to get a face bounding box and I’m getting “Failed to load feature: The file could not be found” error.
This is on a Windows 10 machine with an NVIDIA RTX 3070. I’ve also installed the NVIDIA Broadcast AR SDK. Did I miss something in the setup, or might this be an actual bug?
I’ve also followed the steps to create a face mesh, which worked great, but doesn’t help this CHOP get past this error.
Are you getting that error message with just the chop in your network or do you have the sop there as well? Having the sop linked will require the mesh file even for just bounding boxes.
I can’t reproduce the issue yet on my machine, but we’re looking into it. Have you updated your nvidia graphics drivers recently? I believe it requires at least 455 or later.
Hello Greg,
I tried your .toe but it doesnt work “failed to lod features…”
I tried also to use the command prompt to create the nvf file but it also failed:
It indicate the file “sfm_shape_3448.bin” cannot be opened.
All the indicated file are in the same folder as "Convert…Model.exe
What is wrong?
G:\TD\face>ConvertSurreyFaceModel.exe --shape=path/sfm_shape_3448.bin --blend_shape=path/expression_blendshapes_3448.bin --topology=path/sfm_3448_edge_topology.json --contours=path/sfm_model_contours.json --ibug=path/ibug_to_sfm.txt --out=output-path/face_model0.nvf
“path/sfm_shape_3448.bin”: The file could not be opened
ConvertSurreyFaceModel [flags]
where flags are:
–verbose[={true|false}] produce verbose output
–debug[={true|false}] write extra debugging information
–shape=<file.bin>|<file.nvf> specify the input EOS shape file
–blend_shapes=<file.bin> specify the input EOS blend shape file
–topology=<file.json> specify the input EOS topology file
–contours=<file.json> specify the input EOS contours file
–ibug=<file.txt> specify the input EOS ibug mapping file
–in= specify the input file *.nvf
–out= specify the output file *.nvf or *.json
–recompute_topology[={true|false}] recompute the topology
ERROR: The file could not be opened
Finaly I was able to generate the .nvf file, suppressing all the “path/” in the text from the wiki.
But is no more working…
Trying to update to the last build?
still no success (RTX3090)
here is the working prompt
ConvertSurreyFaceModel.exe --shape=sfm_shape_3448.bin --blend_shape=expression_blendshapes_3448.bin --topology=sfm_3448_edge_topology.json --contours=sfm_model_contours.json --ibug=ibug_to_sfm.txt --out=face_model0.nvf
Sorry for the confusion, I just copied those instructions from the Nvidia documentation. Path was just meant as a placeholder to whatever location those files are in. If they are in the same folder as the exe then you should just list the filename. I’ll update the docs to try and make that clearer.
We’re still looking into the “failed to load features” issue. So far it seems to be working on the machines that we have here available to test, so I’m not sure what the problem is.
As a separate note, we did just identify a problem with the 126 pt landmark mode that we’re fixing at the moment (there appear to be some incompatibility with the landmark feature settings and mesh fitting that we’ll need to add some warnings for).
Hi guys, i’ve tried to make work on my machine the last face track chop… i did firstly all the steps, (nothing is worked for me) and after i’ve asked help using other face_model0.nvf from other users… in this case too no way…
my configuration is the amd 5950x and 3090… latest cuda installed, ar sdk installed.
Suggestions?
test with the default model path (error - failed to load feature: the file could not be found)
A quick update for everyone: Nvidia just release an update for their face tracking system (now part of Nvidia Maxine) which seems to fix the “file not found” issue that was happening on 30xx series cards. We’ve still got a few things to finish testing with the update, but I’ll post an unofficial build you can try as soon as its ready.
It uses the new 0.7.5 release of Nvidia’s AR library which seems to work properly now on our 30xx series test system. As with other unofficial builds, this hasn’t gone through our regular testing procedure, so there may be other issues.
